Any tips to improve the performance of spectral-norm in benchmarksgames

I am trying to implement the spectral-norm benchmark game and below is my Raku implementation. It closely follows the Python implementation here. It produces correct results, however, it is so slow. For example, when the input is 200, Python runs around 1.3sec while Raku takes around 6.4sec in my laptop. Anyone has any good practices to improve its performance?

sub A(\i, \j) {
  1e0 / ((i+j)*(i+j+1e0)/2e0 + i + 1e0)
}

sub Av(@v, \i) {
  [+] ( -> \j { A(i, j) *  @v[j] } for ^@v.elems );
}

sub Atv(@v, \i) {
  [+] ( -> \j { A(j, i) *  @v[j] } for ^@v.elems );
}

my $*SCHEDULER = ThreadPoolScheduler.new(:max_threads(4));
sub AtAv (@v) {
  my @u-promise = (^@v.elems).map(-> \i { start Av(@v, i) });
  my @u = await @u-promise;
  my @w-promise = (^@u.elems).map(-> \i { start Atv(@u, i) });
  await @w-promise;
}

multi sub MAIN(Int $n) {
  my @u := (1e0 xx $n).list;
  my @v;

  for ^10 {
    @v := AtAv(@u);
    @u := AtAv(@v);
  }

  my $a = start { [+] @u [Z*] @v };
  my $b = start { [+] @v [Z*] @v };
  printf "%.9f", ($a.result / $b.result).sqrt; 
}